Distribution Network Topology Control Using Attention Mechanism-Based Deep Reinforcement Learning

Abstract

As the distributed energy mainly based on wind and solar energy continues to be incorporated into the power grid, its automatic control and management has become a very complicated task, and it needs to seek more intelligent control technology. In this paper, a deep reinforcement learning method SAC (Soft Actor-Critic) combined with attention mechanism is proposed to manage power grid. This method changes the line connection and bus distribution of the substation by adjusting the topology structure of the power grid, so that it can transmit power efficiently. And by assigning different feature weights, the attention mechanism enables the neural network to focus on the input that is more relevant to the current target task from a large number of grid input feature states, which enhances the robustness and computational efficiency of the model. And Experiments have proved that our algorithm can automatically manage three different size distribution networks IEEE-5, IEEE-14 and L2RPN WCCI 2020 for three days without experts' help and make sure them run properly and safely.